One reviewer per layer: splitting code review by the way code is read
Summary
This article advocates a layered approach to code review driven by AI-assisted reviewers. It describes six review layers (Intent, Form, Local meaning, Wiring, Runtime, Use) and four lenses (Correctness, Security, History, Reversibility), organized as a tree of agents that read code at different abstraction levels. It also references historical modular design concepts and suggests testing the pipeline by labeling incidents to map where bugs occur.